[
https://issues.apache.org/jira/browse/HDFS-1580?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13010767#comment-13010767
]
Ivan Kelly commented on HDFS-1580:
----------------------------------
@Todd
Ah ok, since rolling has to be exposed, we should expose it fully. Your segment
suggestion sounds good. It will require a separate array in FSEditLog though,
one for JournalManagers, one for EditLogOutputStreams. I'll have to think about
this. Perhaps we could get rid of the journal manager abstraction completely
and extend EditLogOutputStream to handle rolling, as rolling is the only
operation that changes for output with transactions.
Then we could have a separate class that handles stuff like format and getting
a list of the logs. I'll think about this more and put up a new design.
@Jitendra
1. I think I mentioned the second alternative earlier in the thread, but yes,
#transfer will go away.
2. A single bookkeeper setup can be shared by many writers. The more writers
you have the slower the reads, as ledgers are interleaved when written to disk.
3. There's two layouts to think about, the storage layout and the data layout.
The first will be entirely internal to the JournalManager. The second will need
to be saved though.
4. Very good point.
5. Ok
6. Will do
> Add interface for generic Write Ahead Logging mechanisms
> --------------------------------------------------------
>
> Key: HDFS-1580
> URL: https://issues.apache.org/jira/browse/HDFS-1580
> Project: Hadoop HDFS
> Issue Type: Improvement
> Reporter: Ivan Kelly
> Attachments: HDFS-1580+1521.diff, HDFS-1580.diff,
> generic_wal_iface.pdf, generic_wal_iface.pdf, generic_wal_iface.txt
>
>
--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira
